✨ One suite, ready to install

This repository is the independently installable context-economy distribution from TIKAZ AI Skills for Codex. The monorepo is the canonical development source; this repository is automatically synchronized and optimized for people who need only this workflow.

Designed, integrated, refactored, and continuously maintained by TIKAZ; community contributions are welcome. This is an independent project, not an OpenAI-official repository.

This repository is an automatically synchronized distribution of the canonical TIKAZ-AI-Skills monorepo. Cross-suite issues and source changes belong in the canonical repository.

🧩 What makes it different

  • Paste text, attach files, or provide a webpage and let the workflow choose Text / Hybrid / Source automatically
  • The optional pinned Defuddle adapter keeps source HTML, cleaned HTML, Markdown, metadata, protected facts, and separate byte/token estimates
  • Public-URL safety checks, bounded fetching, pending visual evidence, and source-preserving fallback keep uncertain pages visible instead of pretending extraction succeeded

📦 Install

Install the dependency-free core from GitHub. pipx is recommended because it keeps the command isolated:

pipx install git+https://github.com/TIKAZI/TIKAZ-Codex-Context-Economy.git

Fallback for the active Python environment:

python -m pip install git+https://github.com/TIKAZI/TIKAZ-Codex-Context-Economy.git

tikaz-context doctor
tikaz-context pack --input notes.md --query "prepare release evidence" --budget 800 --output .context-economy
tikaz-context benchmark --output .context-benchmark

Uninstall the isolated CLI with pipx uninstall tikaz-context-economy. To install the Codex workflow rather than the CLI, clone or download this repository and copy the root orchestrator or one child Skill into the Skill directory supported by your host.

🧩 Use one Skill independently

Every Skill below has its own promise, installation command, example, execution contract, limits, and bilingual project page:

🚀 Try it

Process my pasted text, attached files, or webpage for this task. Automatically choose Text, Hybrid, or Source, prepare reusable Markdown where safe, retain protected facts and anchors, and make every omission visible.

Extract this webpage into traceable Markdown. Keep the source HTML, metadata, images and complex-table warnings; do not treat pending visuals as inspected.

Create a recoverable checkpoint from this long conversation. Preserve decisions, constraints, completed evidence, file paths, numbers, commands, and open questions.

📊 Evidence status

  • Measured: 50 checked-in synthetic cases, including six long-context variants with 69.7% estimated reduction; declared protected facts and expected anchors are reported separately.
  • Pending: provider billing telemetry, real-world and scanned-PDF generalization, vision-description accuracy, and downstream blind-answer quality.
  • Unavailable until verified: external adoption counts and testimonials. No usage claim is inferred from Stars or anonymous feedback.

Run tikaz-context benchmark --output .context-benchmark and inspect the raw cases, metrics, and summary before repeating any number as a general claim.

🔐 Privacy, security, and community evidence

The core runs locally and does not automatically upload inputs, generated packs, diagnostics, or usage telemetry. Optional converters and the Defuddle adapter are explicit external boundaries. Read the threat model and security policy before processing untrusted sources.

Used it on a public or sanitized task? Submit a verifiable user story with the version, input profile, command, before/after measurements, fidelity checks, and reproducible artifacts. Unverified submissions are not promoted as adoption evidence.

🔄 How the suite works

Read SKILL.md for the owning workflow, references/routing.md for specialist routing, and references/output-contract.md for the verified handoff. Optional tools are detected at runtime; local login state or machine-specific software is never promised as universally available.

🗂️ Repository structure

./
├─ SKILL.md                 # suite orchestrator
├─ agents/                  # Codex UI metadata
├─ references/              # routing and output contract
├─ <child-skill>/           # independently installable specialists
├─ DISTRIBUTION.yml         # canonical source and sync metadata
└─ scripts/                 # deterministic validation

⚖️ Canonical source and contributions

Development, source review, and cross-suite architecture live in TIKAZ-AI-Skills. This distribution synchronizes from suites/context-economy every week and can also be refreshed manually through GitHub Actions.

Source modes, observed licenses, and concrete TIKAZ contributions are recorded in SOURCES.yml and THIRD_PARTY_NOTICES.md. TIKAZ-authored files are released under the MIT License.

🌐 Explore the TIKAZ workflow family

🏠 AI Skills · ⚡ Context Economy · 🎨 Frontend Design · 🎬 Video Intelligence · 🛠️ Engineering · 🔬 Research · 📽️ Presentation · 🖼️ Visual Content